Abstract:

There will be disadvantages like 2-pixel slash thinning distortion, 2×2 square structure loss and the existence of a large number of diagonal redundant pixels when ZS thinning algorithm is used to refine the target image. Furthermore, the mainstream skeleton extraction algorithm can not solve the marginal branching problem caused by unsmooth contour. Based on ZS thinning algorithm, the smooth iterative process and the subsequent scanning process are introduced, the judgment of template retention and template deletion conditions are added in this paper. Experimental result shows that the improved thinning algorithm can keep the 2-pixel slash and square structure far from losing, completely delete redundant pixels on the basis of preserving the skeleton information and topological properties of the target image. The thinning rate of the improved algorithm is 0.05% - 0.25% higher than ZS, IEPTA and MZS algorithms. Meanwhile, the increase of smoothing iteration times can further improve the thinning degree, reduce a lot of marginal branches and improve the smoothness of the overall contour.

Key words: skeleton extraction algorithm, smoothing iteration, template matching, marginal branches

摘要：

在使用ZS细化算法对目标图像细化时，会出现二像素宽度斜线结构细化畸变、2×2正方形结构丢失，以及大量斜线冗余像素存在的弊端，同时主流骨架提取算法无法解决不平滑轮廓带来的边缘分叉问题。针对四类问题，在ZS细化算法基础上引入了平滑迭代流程以及后续的扫描过程，并在其中加入保留模板和删除模板条件的判定。实验数据表明，改进算法在保留目标图像的骨架信息和拓扑性质的基础上，能保持二像素宽度斜线和正方形结构不丢失，并完全删除冗余像素，其细化率相比ZS、IEPTA、MZS细化算法提高了0.05%~0.25%不等。同时平滑迭代次数的增加，能进一步提高细化程度，减少大量的边缘分叉并提高整体轮廓的平滑程度。

关键词: 骨架提取算法, 平滑迭代, 模板匹配, 边缘分叉